To separate the discussion about the optimization-strategy api for v5 started in #101.
In the future it makes sense to rename this feature to something like "CompositeOptimizer" or "OptimizerPipeline".
In scikit-learn the estimators are passed to the Pipeline as a list of estimators. This could be the way to go in Hyperactive as well. Just passing a list of Optimizers to a Composite-class.
An alternative approach would be a builder pattern, like:
comp = CompositeOptimizer(...)
comp.add_optimizer(RandomSearch)
comp.add_other_stuff()
comp.add_condition()
comp.add_optimizer(HillClimbing)
This would make sense if we have another things as optimizers, that we want to add (in the future). Maybe additional conditions, like if-else conditions, that decide which optimizer to choose next.
To separate the discussion about the optimization-strategy api for v5 started in #101.
In the future it makes sense to rename this feature to something like "CompositeOptimizer" or "OptimizerPipeline".
In scikit-learn the estimators are passed to the Pipeline as a list of estimators. This could be the way to go in Hyperactive as well. Just passing a list of Optimizers to a Composite-class.
An alternative approach would be a builder pattern, like:
This would make sense if we have another things as optimizers, that we want to add (in the future). Maybe additional conditions, like if-else conditions, that decide which optimizer to choose next.